tevador: can’t get the asm code to compile for iOS

always syntax issues

though I might be doing something wrong

selsta: it looks like you are compiling for ARMv7, but your cmake is configured for x86_64

tevador: I used this: cmake.org/cmake/help/latest/manual/…s-compiling-for-ios-tvos-or-watchos

can’t figure it out, I will just set the ifdef to macOS only

try to add -DCMAKE_SYSTEM_PROCESSOR=aarch64

looks better now: paste.debian.net/hidden/a811b06b

yeah, __clear_cache doesn't work on Apple, you should add something like #ifndef APPLE

it compiled now at least

although if Macbooks also start using ARM, they will need some function to flush the cache or JIT won't work

problem is I can’t install it on my phone without xcode

and xcode has issues with the asm syntax again

lol I give up

best chance to test the arm stuff is once someone has an arm mac

why can't you just compile the asm code separately and then link the .o files into the xcode build?

I bet xcode will fail to link it

because it won't have Apple (c) (tm) in the header or something

there's gotta be a binutil that can construct the necessary header, sheesh

still would be interesting to see why xcode fails in the first place. is it trying to build fat binaries, and assemble to some other platform than the asm file was meant for?

I’m sure it is possible, there are monero ios wallets

though frankly, I don't think it's much loss if we drop support for Apple

even M$ is more open-source friendly these days

same error inside xcode

no....

I meant, copy the .o files from the asm code built by gcc into the xcode build tree so that it doesn't try to compile/assemble them again

the other possibility is that xcode clang needs a different suffix than ".S" to recognize a file as asm source

or the actual invocation of clang is forcing it to treat that source file as C instead of asm

where did RandomX.xcodeproj come from?

cmake generated it itself?

yes, cmake can generate xcode projects

well, I would invoke that last failed command manually

and drop most of the C-related flags

in particular, "-x c"

hyc: yep, removing -x c seems to work

this talks about explicitly setting the file type, maybe this has to be set to ASM

yes, but as I recall that broke something else

teavdor would know more about it

tevador

ok, nice

I can explicitly set it to assembly inside xcode

now I should be able to run it on my phone

see the git history for CMakeLists.txt and the comment "cheat because cmake and ccache hate each other"

note that there's already a special case for the asm on MSVC, you can probably do something similar for xcode

will look into this later

I also had to manually remove -latomic form the linker flags

for it to compile

> Performance: 1323.39 ms per hash

lol

anything I should test? seems like I can’t use --jit else it will crash

and for --mine mode I don’t have enough RAM

this is on your phone?

yes

I thought we set the proper mmap flags to let it use JIT

afaik JIT is not allowed on iOS

at all? what are they doing with webasm?

iOS forbids JIT by design

unless your phone is jailbroken

you need private entitlements to run JIT and only Apple can use those

selsta: have you figured out how to detect iOS?

sounds like ios is all more trouble than it's worth

we could disable JIT there

what's going to happen when the ARM macbooks are out, and they tell you you can run all your apple software on phone or laptop directly

when only using `--verify`, adding MAP_JIT to mmap does not make a difference

when using with --jit, adding MAP_JIT results in cache allocation failed, without MAP_JIT is crashes with bad access

is this macos or ios?

iOS

what does it do on macos?

adding MAP_JIT?

yes

it allows you to enable hardened runtime with the `com.apple.security.cs.allow-jit` entitlement

without it you have to use `com.apple.security.cs.disable-executable-page-protection` entitlement which is "less secure"

W^X is useless anyways unless you also have some protection against ROP

i.e. injected return addresses on the stack

don't they also do ASLR?

difficult to do anything useful with ROP unless you can find a useful call in the running binary itself

not having __clear_cache() on iOS is probably why it'll crash with JIT anyway

ah right I disabled that

well, there must be a way to clear the cache because safari uses JIT

well, I wrote this for xmrig: github.com/xmrig/xmrig/blob/master/…o/randomx/jit_compiler_a64.cpp#L110

but I'm not sure if it works on iOS at all

pointless to pursue it

iOS processes only have 2GB address space, so can never be effective miners

Thread 1: EXC_BAD_ACCESS (code=2, address=0x10607c000)

largest mmap'able region is less than 1GB

that is with sys_icache_invalidate

but yea agree kinda pointless, we will have to wait for arm macs

iOS address space is a recurring problem for LMDB users

funny that the latest iphones still only have 4GB RAM

my phone is 4 or 5 years old lol

don’t even know if it has more than 1 gb ram

ok it has 2gb ram

iPad pro 11 has 6GB woohoo

anybody doing ipad benchmarks?

well at least I was able to run randomx-benchmark on my phone :D just super slow

only in light mode hh

I wonder how iOS wallets do randomx verification

they certainly are faster than 1 hash/s

wallets don't do verification

?

ah right :D

not sure why I thought that they do block verification

wow, nanopool is #1 now

more cloudnets?

new one

the other two are still there

I wonder who is paying for all this

if it’s someone abusing free credits then they would have fixed it by now

0.325GH total, 20% of the network